Comparing classical pathways and modern networks: towards the development of an edge ontology

Long Lu, Andrea Sboner, Yuanpeng J. Huang, Hao Lu, Tara A. Gianoulis, Kevin Y. Yip, Philip M. Kim, G.t. Montelione, Mark Gerstein

This paper, published by researchers at Yale and Rutgers, addresses a fundamental tension in systems biology: the gap between the "classical" diagrams found in textbooks and the "modern" hairballs generated by high-throughput data.

TL;DR

The authors propose a framework to "embed" classical biological pathways into large-scale interaction networks, revealing hidden crosstalk and isoform-specific details. To bridge the gap between mechanistic detail and big-data scalability, they introduce a prototype Edge Ontology—a standardized language for describing the diverse relationships between biomolecules.

What Problem Does This Paper Address?

In biology, we have two ways of looking at the world. Classical pathways (like the Citric Acid Cycle or Notch signaling) are mechanistic and intuitive, but they are often "reductionist" islands that ignore how pathways bleed into one another. Furthermore, the arrows in these diagrams are ambiguous; an arrow might represent a physical movement, a chemical transformation, or an abstract activation.

On the other hand, modern networks (derived from high-throughput experiments) are mathematically rigorous and capture the "big picture," but they are often "shallow." They treat every interaction as a simple line, losing the rich biochemical context of how two proteins interact.

The authors argue that we lack a standardized "Edge Ontology"—a way to define what those lines and arrows actually mean—which prevents us from effectively merging these two perspectives.

Methods

The researchers employed a multi-step approach to unify these views:

  1. Pathway Embedding: They took "core components" from classical databases (like KEGG) and mapped them onto a massive Human Protein-Protein Interaction (PPI) network.
  2. Extended Embedding: They expanded these pathways to include "first neighbors"—proteins that interact with the core components but aren't officially part of the "textbook" pathway.
  3. Topological Analysis: They used graph theory metrics (like average degree, betweenness, and clustering coefficients) to see if different types of pathways (e.g., metabolic vs. signaling) have distinct mathematical "shapes."
  4. Ontology Design: They developed a four-level hierarchical ontology for edges, using specific shapes, colors, and symbols to represent different biological actions (e.g., phosphorylation, cleavage, or translocation).

Datasets

The study utilized several key biological databases:

  • KEGG & BioCarta: Used as the source for "classical" core pathway components.
  • HPRD (Human Protein Reference Database): Used as the large-scale "modern" network for embedding.
  • EcoCyc, WIT, and MetaCyc: Referenced for metabolic pathway comparisons.

Key Insights and Ideas

  • Crosstalk is the Rule, Not the Exception: While classical pathways look isolated, the "extended" embedded pathways show massive overlap. In fact, the authors found that almost all human signaling pathways can be connected to one another by just one degree of separation.
  • Topology Reflects Function: Signaling pathways and metabolic pathways aren't just biologically different; they are mathematically different. Signaling pathways tend to have higher "betweenness" (acting as information conduits) and higher average degrees (more connections) than metabolic pathways.
  • Isoform Specificity Matters: Classical diagrams often use one box for "Notch." By embedding this in a real-world network, the authors showed that only one specific version (Notch1) interacts with certain regulators like Numb, a detail usually lost in simplified diagrams.
  • The Need for an "Edge" Language: We already have the "Gene Ontology" (GO) to standardize how we talk about nodes (proteins/genes). This paper argues we desperately need a similar standard for edges (interactions) to make high-throughput data computationally "readable" without losing mechanistic nuance.

Key Results

  • New Players: The extended embedding of the Notch pathway identified 218 new proteins potentially involved in the process that were missing from the KEGG core diagram.
  • Clustering Success: Using only five network statistics, the authors could successfully cluster pathways by their biological category. For example, 88% of metabolic pathways grouped together based on their unique topological signature.
  • The Ontology Prototype: The authors presented a 4-level hierarchy (Table 1) that classifies interactions from general (Directed vs. Undirected) to specific (e.g., N-linked glycosylation). They demonstrated its utility by redrawing the Notch and JAK-STAT pathways, making them both mathematically precise and biochemically rich.

Limitations and Future Work

The authors acknowledge that their proposed edge ontology is a "prototype" and far from complete; building a full version will require years of community effort. A significant limitation is the lack of temporal information—current network models are static "snapshots" and don't yet capture the timing of biological events. The next "grand challenge" will be incorporating this explicit time-dimension into the edge ontology to model how pathways pulse and flow in real-time.

Summary updated

Abstract

Pathways are integral to systems biology. Their classical representation has proven useful but is inconsistent in the meaning assigned to each arrow (or edge) and inadvertently implies the isolation of one pathway from another. Conversely, modern high-throughput experiments give rise to standardized networks facilitating topological calculations. Combining these perspectives, we can embed classical pathways within large-scale networks and thus demonstrate the crosstalk between them. As more diverse types of high-throughput data become available, we can effectively merge both perspectives, embedding pathways simultaneously in multiple networks. However, the original problem still remains - the current edge representation is inadequate to accurately convey all the information in pathways. Therefore, we suggest that a standardized, well-defined, edge ontology is necessary and propose a prototype here, as a starting point for reaching this goal.
BibTeX
@inproceedings{lu2007comparing,
  author = {Long Lu and Andrea Sboner and Yuanpeng J. Huang and Hao Lu and Tara A. Gianoulis and Kevin Y. Yip and Philip M. Kim and G.T. Montelione and Mark Gerstein},
  title = {{Comparing classical pathways and modern networks: towards the development of an edge ontology}},
  booktitle = {Trends},
  year = {2007},
  doi = {10.1016/j.tibs.2007.06.003},
}

Loading related content...

Figure